Let $z_x$ and $z_y$ represent the standardized versions (z-scores) of variables $x$ and $y$. If the least-squares regression line for predicting $z_y$ from $z_x$ has slope $m_1$, and the least-squares regression line for predicting $z_x$ from $z_y$ has slope $m_2$, which of the following is true for any non-zero, non-perfect correlation?